Journal of Hydrology 367 (2009) 70–78

Contents lists available at ScienceDirect

Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol

A compilation of data on European flash floods

Eric Gaume a,*, Valerie Bain b, Pietro Bernardara c, Olivier Newinger a, Mihai Barbuc d, Allen Bateman e,
Lotta Blaškovičová f, Günter Blöschl g, Marco Borga h, Alexandru Dumitrescu i, Ioannis Daliakopoulos k,
Joachim Garcia e, Anisoara Irimescu i, Silvia Kohnova j, Aristeidis Koutroulis k, Lorenzo Marchi l,
Simona Matreata d, Vicente Medina e, Emanuele Preciso l, Daniel Sempere-Torres m, Gheorghe Stancalie i,
Jan Szolgay j, Ioannis Tsanis k, David Velasco m, Alberto Viglione g
a
Laboratoire Central des Ponts et Chaussées, Nantes, France
b
CEREVE, ENPC, Ecole des Ponts Paris Tech, Université Paris Est, 6-8 Avenue Blaise Pascal, Champs-sur-Marne, 77455 Marne-la-Valée, France
c
LNHE, EDF R&D, quai Watier, 78401 Chatou, France
d
National Institute of Hydrology and Water Management, Romania
e
UPC - GITS (Sediment Transport Research Group), Universitat Politécnica de Catalunya, 08034 Barcelona, Spain
f
Slovak Hydrometeorological Institute, Bratislava, Slovak Republic
g
Vienna University of Technology, Karlsplatz 13/222, A-1040 Wien, Austria
h
University of Padova, 35122 Padova, Italy
i
National Meteorological Administration, 97, Soseaua Bucuresti-Ploiesti, Sector 1, 013686 Bucharest, Romania
j
Slovak University of Technology Bratislava, Radlinskeho 11, 813 68 Bratislava, Slovak Republic
k
Technical University of Crete, Laboratory of Water Resources Management and Coastal Engineering, Chania 73100, Greece
l
CNR IRPI, Corso Stati Uniti 4, 35127 Padova, Italy
m
Grup de Recerca Aplicada en Hydrometeorologia (GRAHI), Universitat Politécnica de Catalunya, 08034 Barcelona, Spain

a r t i c l e i n f o s u m m a r y

Article history: Flash floods are one of the most significant natural hazards in Europe, causing serious risk to life and
Received 16 June 2008 destruction of buildings and infrastructure. This type of flood, often affecting ungauged watersheds,
Received in revised form 11 December 2008 remains nevertheless a poorly documented phenomenon. To address the gap in available information,
Accepted 30 December 2008
and particularly to assess the possible ranges for peak discharges on watersheds with area smaller than
500 km2 and to describe the geography of the hazard across Europe, an intensive data compilation has
This manuscript was handled by K. been carried out for seven European hydrometeorological regions. This inventory is the first step towards
Georgakakos, Editor-in-Chief, with the an atlas of extreme flash floods in Europe. It contains over 550 documented events. This paper aims at
assistance of Demetris Koutsoyiannis, presenting the data compilation strategy, the content of the elaborated data base and some preliminary
Associate Editor. data analysis results. The initial observations show that the most extreme flash floods are greater in mag-
nitude in the Mediterranean countries than in the inner continental countries and that there is a strong
Keywords: seasonality to flash flood occurrence revealing different climatic forcing mechanisms in each region.
Floods Ó 2009 Elsevier B.V. All rights reserved.
Flash flood
Extremes
Database
Atlas
Regional analysis

Introduction
Flash floods are one of the most significant natural hazards and
cause serious loss of life and economic damage. The average annual
economic loss due to natural hazards over the world has been esti-
mated at €40 billion (Münich Re, 2003). This can be compared to
the total economic damages estimated at €1.2 billion for the Gard
2002 single flash flood event (Huet et al., 2003) and €3.3 billion for
* Corresponding author. Tel.: +33 1 40845884; fax: +33 2 40845998.
E-mail address: gaume@lcpc.fr (E. Gaume).
the Aude 1999 flash flood (Lefrou et al., 2000). In Europe, lowland
floods are rarely associated with fatalities except in cases of levee
failures; in contrast flash floods often result in loss of life. The most
striking examples are the Lynmouth flood in the UK in 1952 – 34
deaths (Dobbie and Wolf, 1953), the Barcelona flood in Spain in
1962 – over 400 deaths (López Bustos, 1964), the Piedmont region
floods in Italy in 1968 and 1994 – respectively, 72 and 69 deaths
(Ferro, 2005; Guzzetti et al., 2005) and the Aude flood in France
in 1999 – 35 victims (Gaume et al., 2004).
Despite being a serious natural hazard that affects countries
throughout Europe, flash floods remain a poorly understood and

0022-1694/$ - see front matter Ó 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.jhydrol.2008.12.028
 E. Gaume et al. / Journal of Hydrology 367 (2009) 70–78
documented natural phenomenon. The occurrence of extreme
events in ungauged watersheds generally means that there is no
measured discharge information or formal records of the magni-
tude of the event. Furthermore, data on previous flash floods is scat-
tered among local authorities where flooding has occurred and
various companies and research units that may have unpublished
technical reports. These sources of information are often difficult
to obtain and are generally in the national language of the country
where the flood occurred. Existing inventories of past floods (Bar-
redo, 2007; Herschy, 2005; Costa, 1987a; Unesco, 1976; Pardé,
1961) contain only few data on flash floods occurred in Europe
and lack quantitative information on the meteo-hydrological and
hydraulic characteristics of the events. The lack of centralised na-
tional and international databases for storing meteo-hydrological,
hydraulic and socio-economic data relating to past flash flood
events within Europe hinders the development of understanding
of their magnitude and occurrence (Creutin and Borga, 2003).
To address the gap in available information, and particularly to
assess the possible ranges for peak discharges on small watersheds
and to describe the geography of the hazard across Europe, an
intensive data compilation has been carried out initially for the se-
ven European regions listed below:
 Catalonia region, Spain, Mediterranean region,
 Cevennes-Vivarais, France, Mediterranean region,
 Italian Alps and Ligury, Alpine Mediterranean and Mediterra-
nean region,
 Slovakia, Inland Continental region,
 Greece, Mediterranean region,
 Romania, Inland Continental region,
 Austria, Alpine Inland Continental region.
This inventory is the first step towards an atlas of extreme flash
floods over Europe. This paper presents the data collection strategy,
the data set and some first interpretations. This work has been car-
ried out for Hydrate, which is a currently ongoing European Commis-
sion funded project that is aiming to improve techniques for flash
flood forecasting. Such inventories can never be perfectly consistent
and comprehensive. The very first analyses provided hereafter nev-
ertheless illustrate their usefulness for understanding magnitude,
occurrence, and geographical distribution of hydrological extremes.
Data collection strategy
The initial aim of the data compilation was to develop a cata-
logue for each region that included the most extreme flash flood
events between 1946 and 2007. In this research, extreme flood
events induced by severe stationary storms have been considered
as flash floods. This relatively broad definition includes almost all
the past events reported as flash floods in Europe, except dam
break floods. The duration and spatial extension of the area af-
fected by such floods depend on the causative storm and hence
on the climatic setting. Most generally, the storms inducing flash
floods lead to local rainfall accumulations exceeding 100 mm over
a few hours and affect limited areas: some tens to some hundreds
of square kilometres. Larger scale and longer lasting stationary
storm events may, however, occur in some meteorological con-
texts, especially in the Mediterranean region. As an example, an
area larger than 3000 km2 received more than 300 mm rainfall
within about 12 h on the 8th and 9th of September 2002 in the
Gard region of France (Delrieu et al., 2004). On the basis of these
considerations, it has been decided that the most extreme floods
in watersheds of an area of less than 500 km2, generally induced
by short duration storms (i.e. less than 24 h) should be considered
as flash floods. Matthai (1969) and Stanescu (2000, 2004) used a
similar watershed area threshold when compiling data on flash
71
floods. This initial criterion was, however, relaxed for several cases
of floods in France and Catalonia which occurred over watersheds
of much greater areas but which were created by short duration
and very high accumulation storms. Examples are the Gard 2002
floods in France that has already been mentioned; other large scale
flash floods include events on the Llobregat River in Catalonia in
1971, the Segre River in Catalonia in 1982, and the Ardèche River
in France in 1982, 1995 and 1996, reported in the Hydrate data
base.
The objective was to document a minimum number of 30
floods in each region: if possible the events considered as the
most extreme or ‘‘top 30” flash floods and homogeneously distrib-
uted over the selected period. The number of described events
was chosen to build representative samples of large flash floods
in the various regions, not limited to the very extreme events.
The timeframe was selected based on the hypotheses that firstly,
over a 60 year period it would be likely that several extreme
events would have occurred in each region; and secondly, that
it would be feasible to find hydrometeorological data from this
period but that extending the record to earlier dates, with the
objective of building an inventory as exhaustive as possible,
may not be possible for most regions.
Data was collated by local researchers in each region. Several
different types of data sources were used to identify the dates
and location of extreme flood events and gather detailed informa-
tion on these events. Typically, information came from available
discharge and rainfall records, scientific and technical reports, local
flood risk mapping studies and site investigation data.
Data was collated using a standardised template containing sec-
tions on geographic, meteorological, hydrological and hydraulic
data, as well as information on damages and casualties that were
caused by the flood. The content of the data catalogue will be de-
scribed in more detail in the next section. As much as possible infor-
mation was collated on the considered flood event and if possible the
sources of documentation were scanned and appended to the tem-
plates. The minimum requirement for an event to be listed in the
database is that at least one peak discharge estimation for one given
cross-section is available and that the corresponding watershed area
could be computed (i.e. the cross-section could be located).
Validation of the datasets was carried out, when possible, by
cross-checking reported rainfall accumulations with peak dis-
charge estimates and identifying outliers from the rest of the data-
set for the region. Overall, intraregional comparisons and
comparisons with other inventories did not reveal significant dis-
cordances in peak discharge estimates except for isolated events.
For example, the reported estimated peak discharge for the event
on the Rubí Torrent in Catalonia in 1962 was thought to be dubi-
ous, because the resulting unit peak discharge (72 m3/s/km2) is
considerably higher than any of the other reported unit peak dis-
charges in the inventory for similar watershed areas and indeed
higher than the world envelope curve. This led to a re-analysis of
the cross-section surveys realized just after this flood and of the
initial peak discharge estimates (López Bustos, 1964). The peak dis-
charge of 1750 m3/s corresponds to a river cross-section area of
about 100 m2, which means an average flow velocity of more than
17 m/s. Velocities are hardly greater than 4–5 m/s for liquid flows
in natural channels with shallow slopes, typically slopes lower
than 2% (Gaume et al., 2004; Jarrett, 1987). The discharge is clearly
largely over-estimated. Many reasons can be put forward to ex-
plain such an error and a more accurate guess would deserve a
thorough analysis and the compilation of additional field data: pic-
tures or accounts that could help to evaluate the possible range of
flow velocities and their repartitions in the considered cross-sec-
tions for instance. The 1962 event is not included in the initial data
set, however, after a critical re-analysis of the event is completed, it
will be added to the data catalogue.
72 E. Gaume et al. / Journal of Hydrology 367 (2009) 70–78

Table 1
Content of the hydrate flash flood database (required data for all events in normal style and optional additional data in italics).

Section identification
Data
Event code
Date of the event
River name
Cross-section name
Section longitude
Section latitude
Section elevation (m)
Attached documents
Comments and notes
Photos
Attached reports
References list
Basin data Discharge data Rainfall data Damages and casualties
Basin area (km2) Peak discharge (m3/s) Total point rainfall (mm) Total damages (€)
Time of concentration (h) Estimation method Rainfall duration (h) Displaced persons
Minimum elevation (m) Estimation quality rate Av. rainfall on the basin (mm) Population affected
Maximal elevation (m) Regulated stream (y/n) No. of raingauges Direct private damages (€)
Average elevation (m) Peak discharge Maxi No of raingauges in the basin Direct public damages (€)
Average basin slope (%) Peak discharge mini Quality of data Indirect damages (€)
Glacial areas (%) 10-year discharge (m3/s) Type of event Origin of the data
Land use 100-year discharge (m3/s) Spatial extent (km2) No of casualties
Soils Sediment processes (y/n) Max. Intens. over Tc (mm/h) No of injured people
Av. soil thickness (m) Flood duration (h) Hailfall (y/n) Medical causes
Geology Initial wetness (wet/dry) Circumstances
Annual precipitation (mm) Timing
Observation period (years) Gender
Age
Location map Flood hydrograph Rainfall map Report on the damages
Geographical doc. Past-historical floods Radar data Report on casualties
Cross-section survey Local IDF curves
Pictures of the section Monthly precipitations

The data catalogue
The data catalogue is composed of a series of filled data tem-
plates. Their content is presented in Table 1. Each record in the
inventory provides data on the flood characteristics at a particular
cross-section. For any flood event there may be several records,
each with details of the peak discharge at a different cross-sections
and, often, different sub-catchments. The records contain fields
with basic geographic information such as the name of the river,
the name of the cross-section and the longitude and latitude of
the site in WGS84 coordinates for all inventories. There is also a
group of fields with information on the watershed upstream from
the cross-section to which the record relates. Its area should be de-
fined and where possible, an estimation of its time of concentration
is also indicated. Information is given on the elevation and the
average basin slope. For Alpine basins, the proportion of the wa-
tershed which is glacial is estimated. The main land cover for the
watershed is reported. Land cover is recorded following CORINE
land cover definitions from the European Environmental Agency
(http://dataservice.eea.europa.eu). The predominant watershed
soil types and soil thickness is reported as well as the geology.
Maps of soil, geology and topography are appended to the inven-
tory where available.
The discharge data that is given in the inventory aims to provide
details of the specific event that is being documented and also give
some information on the characteristics of previous floods on the
watershed. The event peak discharge (and upper and lower bounds
on the estimate) at the given cross-section is recorded as well as an
indication of the method used for determining it. Most common
methods for the estimation of peak discharges are the reconstruc-
tion by means of indirect methods, the reconstruction from reser-
voir operations and the extrapolation of stage–discharge curves
(Table 2). The quality of discharge estimate is rated into four clas-
ses: (1) Very good; (2) Good; (3) Fair and (4) Poor. Because the
inventories for each region were completed by different organisa-
tions, this quality rating is indicative only since each organisation
used their own metrics to determine the score. Scientific reports
and papers relating to the peak discharge estimation, information
on the flood hydrograph and the cross-section survey are ap-
pended to the inventory. Where possible, the 10 year return period
and 100 year return period discharges for the river at the specific
cross-section are given as well as any specific details that exist
for other flood events on the same watershed.
The inventory also records if needed what the predominant sed-
iment transport processes during the event were. In the scientific
literature, many classifications of water–sediment flows have been
proposed. For the Hydrate project, three classes are proposed and
the event record identifies which of the three best describes the
event:
 Water flood,
 Debris flood or hyperconcentrated flow,
 Debris flow or mudflow.
The term debris flood indicates a process intermediate between
water floods and debris flows. It is similar, although non synony-
mous of hyperconcentrated flow (Pierson and Costa, 1987). A mud-
flow is a debris flow without large clasts. Debris flow and mudflow

Table 2
Methods of estimating the peak discharge of events (%).

Number Manning Strickler Extrapolation of calibrated Hydraulic Hydraulic Reconstruction from Direct current Other Unknown
of records formula estimation stage–discharge relation 1D 2D reservoir operation meter
simulation simulation measurement
Catalonia 10 56 11 33 0 0 0 0 0
France 236 20 33 17 0 7 0 0 23
Italy 73 64 23 13 0 0 0 0 0
Slovakia 52 0 73 0 0 0 0 27 0
Greece 21 66 0 0 17 0 17 0 0
Romania 152 0 53 0 0 47 0 0 0
Austria 34 0 94 6 0 0 0 0 0
 E. Gaume et al. / Journal of Hydrology 367 (2009) 70–78
are non-Newtonian and conventional discharge estimation
methods are not applicable. Different threshold values have been
proposed for differentiating between various types of water–sedi-
ment mixtures. As an example, Costa (1988) indicates sediment
concentration by volume up to 0.2 for water floods with sediment
transport, from 0.2 to 0.47 for hyperconcentrated flows, and from
0.47 to 0.77 for debris flows. Other classifications, however, do ex-
ist. It is well known that, in addition to concentration, particle size
can influence the behaviour of flow: clay and very fine silt can in-
duce non-Newtonian behaviour also in concentrations by volume
lower than 10% (Ning and Zhaohui, 1986). Collecting data on debris
flows is of great interest, but debris flows are a process basically dif-
ferent from water floods and data on debris flow peak discharge are
not homogeneous with water flood data. Quantitative data on deb-
ris flows as estimated discharges, amounts of sediments eroded and
deposited, are only included for a very few number of flash floods
in this inventory as they deserve a separate analysis.
Rainfall data relating to the event are given in a section of the
inventory. The number of rain gauges within and around the wa-
tershed and the number of nearby radar stations as well as the data
for the event and the preceding days are given. When available, the
maximum total accumulated point rainfall is recorded as well as
the spatial extent of the rain, thus enabling an estimate of the aver-
age accumulated rainfall on the watershed. Details of the maxi-
mum intensity, initial watershed wetness and whether there was
hail fall are also noted. The type of hydrometeorological event is
identified as being one of two classes:
 Storm (intense rainfall event),
 Storm on snow.
These two classes define two types of causative processes of
flash floods. Hail fall often occurs during high-intensity storms,
but its influence on the formation of flash floods is not particularly
relevant.
There is a section of the inventory that provides details of the
climatic characteristics of the watershed; however, these fields
are not always complete due to difficulties of obtaining the data.
The fields include details of the average annual precipitation and
the observation period from which the estimate was made, the 1,
10 and 100 year return period hourly and daily rainfall and the
intensity duration frequency curve.
Finally, there are sections within the inventory on damages and
casualties. Estimates (and sources of the estimates) of total eco-
nomic damages, direct damages to private properties and activi-
ties, direct damage to public infrastructure and indirect damages
are given where possible. There are also fields for recording the
number of inhabitants within the affected areas and the number
of displaced people. In terms of casualties, the inventory records
the number of fatalities as well as the number of injured people.
Information on the circumstances of death and injury and the
age and gender of the victims is given. Any available reports on
damages and casualties are appended to the inventory.
Table 3
Number of flash flood events listed in the Hydrate database for each region.
Area (km2) Nb of events Dates
Catalonia 32,000 10 1962–2005
France 18,000 236 1953–2006
Italy 95,000 73 1968–2006
Slovakia 49,000 52 1995–2004
Greece 132,000 21 1960–2006
Romania 240,000 152 1973–2007
Austria 85,000 34 1987–2005
a
Cov: Coverage in yr106 km2.
b
D: density in records/yr/106 km2.
73
Each inventory is completed with relevant additional notes, ap-
pended reports and photographs and a list of references on the
event. Not all records within each regional inventory have success-
fully completed all fields described above but in most cases the
minimum data requirements as specified by the data compilation
template were completed.
The final data set includes 578 flood event records in seven
European regions (Table 3). While the data compilation aimed to
be consistent and comprehensive in each region, the quality and
quantity of data that could be collected varied from country to
country, depending on the available sources of information and
the institutional frameworks for collecting and recording data.
Only in France does the catalogue cover the full 60 year time period
(Fig. 1). Compilation of data on extreme events in France was effi-
cient due to the existence of a systematic program for flood hazard
mapping since 1982 (http://www.prim.net) which has produced
easily accessible information on flood hazard. The peak discharge
computation methods and the corresponding discharge accuracy
rate relate to the nature of the sources of information used. It
can be noted that most of the discharge estimates in the catalogues
for Austria, Slovakia and Romania have been obtained from stage–
discharge relationships, indicating that a large part of the data
comes from gauging stations. In Catalonia, Italy, France and Greece,
many estimates are based on hydraulic calculations, indicating that
a significant proportion of the information is on ungauged sites and
has been collected from scientific reports, papers and studies. In
most regions, the values closest to the envelope curve (see follow-
ing discussion), correspond dominantly to ungauged sites, justify-
ing the efforts for retrieving data from ungauged watersheds.
To improve the consistency of the data and enable first inter-
comparisons, the complete data set has been refined by selecting
the most extreme events, identifying those events which are clos-
est to the envelope curve for each region. The selected envelope
curve approach is presented in the next sections. The 30 events
with the highest ‘‘reduced peak discharge (see next sections for a
definition)” were selected for each region, the reduced discharge
formula being linked to the shape of the envelope curves. Several
of the original data sets contained multiple records for the same
rainfall event, distinguishing different peak discharge estimates
at different locations within the affected catchments. In refining
the data sets to 30 events, it was specified that no more than
two records within the set of 30 should be of the same date in a re-
gion. This condition was selected to maintain a diversity of case
studies and representativeness of the refined data base. It only ap-
plied to the French refined data set which otherwise, due to the
spatial extension of some extreme storm events in this region,
would have had more than half of its records corresponding to
two storm events only: the September 2002 and October 1958
storms. Note that this imposed constraint does not affect the enve-
lope curve (see following comment).
Not all of the refined data sets selected meet the target number of
30 records and the refined data set only counts 150 records over the
seven considered regions (Table 3). The data sets for Catalonia and
Cova Db Refined number Dates Cova Db
1.4 7 9 1971–2005 1.1 8
0.9 255 30 1953–2006 0.9 32
3.6 20 30 1968–2006 3.6 8
0.4 120 30 1995–2004 0.4 70
6.0 3 4 1989–2006 2.2 2
8.1 19 30 1979–2007 6.7 5
1.5 22 17 1987–2005 1.5 11
74 E. Gaume et al. / Journal of Hydrology 367 (2009) 70–78
Fig. 1. Examples of distributions of the reported extreme events for 1947–2007.
Crete, for instance, are small so it will be difficult to draw generali-
sations about floods in these regions or make comparisons with
other datasets. The data set for Austria is also below the target of
30 records and contains no peak discharge estimates on ungauged
watersheds, which will have to be taken into account while making
comparisons with the other sets. Two indices help to compare the
various datasets (reported in Table 3): the compilation coverage
(watershed area multiplied by period of time considered) and the
compilation density (number of documented events divided by
the coverage). The periods of time as well as the areas covered by
the inventories are varied. Let us comment qualitatively on how this
can affect the inter-comparison. The number of observed events
exceeding a given magnitude threshold increases on average line-
arly with the considered area and period of time in a statistically
consistent region if there is a temporal and spatial independence be-
tween events. These two conditions, consistency and independence,
are not fulfilled but the general conclusion remains valid. If we as-
sume that the major floods have been reported in the inventory,
the proportion of high magnitude events in the 30 event sets will
have a general tendency to grow with the coverage of the inventory.
This means that the Italian, Greek and Romanian inventories are
more likely to contain higher magnitude events. In contrast, this
can be partly compensated by the density of the inventories for
France and Slovakia; the proportion of missed extreme events, espe-
cially events that occurred on ungauged watersheds, reduces as the
comprehensiveness of the inventory increases. As a consequence,
the envelope curves are affected by the coverage of the inventories.
The magnitude of the most extreme floods defining the envelope
curve have a general tendency to grow as the coverage grows. Like-
wise, the number of records close to the envelope curve increases
with the density of the inventory.
Having presented the limitations of the available data, the
remaining discussion explores some initial interpretations from
the data analysis, essentially flood peak discharge analysis. Despite
the variability in the quality of each data set resulting in limita-
tions for comparing the data, there are patterns that can be
observed.
First analysis of the resulting data sets
Comparison of envelope curves
Envelope curves provide an effective graphical summary of pre-
vious floods in a given region. They have been widely used in past
publications on extreme floods (Castellarin, 2007; Herschy, 2005;
Bayazit and Onoz, 2004; Stanescu, 2000; Kadoya, 1992; Anselmo,
1985; Mimikou, 1984; Crippen and Bue, 1977; Marchetti, 1953;
Jarvis, 1924) and have the advantage of being relatively unaffected
by the data compilation density because they are determined by
the maximum values of a sample. Generally, the most extreme
flood events in a region have had dramatic consequences except
for floods affecting very small and unpopulated headwater
streams. They remain remembered within communities and are of-
ten well documented. They are therefore easy to identify and count
among the first events reported in regional extreme flood invento-
ries. The maximum discharge values of such datasets do, however,
have a general tendency to grow with the coverage of the inven-
tory: area and period of time considered (Castellarin, 2007). The
highest coverage values do not correspond necessarily to the high-
est number of recorded events (see Table 3).
Envelope curves, adjusted for each region, have been plotted
(Fig. 2). For sake of clarity, a simple envelope curve formula, often
 E. Gaume et al. / Journal of Hydrology 367 (2009) 70–78
Fig. 2. Peak unit discharges of extreme events in the European HOs and envelope
curves.
used in previous studies, has been selected (Eq. (1)). According to
this formula, the envelopes appear as straight lines on the chosen
log–log representation.
Q s ¼ aAb
Here Qs is the unit discharge in (m3/s/km2), A (km2) is the catch-
ment area, a is a coefficient supposed to be independent on the
catchment area also called ‘‘reduced” discharge in (m3/s/km2(1+b)),
and b is a scaling exponent. As suggested by Castellarin (2007),
the value of the exponent b has been estimated through a linear
regression between log (Qs) and log (A) based on each of the refined
data sets. An average value of b = 0.4 appears to be the best suited
to the available data sets. It lies in the range of previously calibrated
envelope curve parameters for various climatic contexts (Castella-
rin, 2007; Jarvis, 1924).
The comparison of the envelope curves shows at least two
groups of curves that appear to correspond to two different cli-
matic influences; Italy, France and Catalonia appear to form one
group and Slovakia and Romania form another. The maximum
peak discharges collated in the regions under Mediterranean cli-
matic influence are more than twice as high as the maximum peak
discharges reported in Central Europe for a given watershed area.
Some very high peak discharge values are observed under the inner
continental climate but they lie far from the maximum values ob-
served in the Mediterranean area. At this stage, no real conclusions
can be drawn for Greece due to the limited number of documented
floods in the database.
The envelope curves for each climatic region are consistent. The
lower position of the Austrian envelope curve remains to be ex-
plained. It may be due to hydro-climatic factors but it may also
be a consequence of the absence of data on ungauged catchments
in this inventory: the inventory coverage is necessarily lower in a
given area if only the gauged catchments are considered. Despite
the high diversity of sources on flash flood discharges used, the re-
trieved values lie in similar ranges. Except for the case of the Rubi
Torrent flood in 1962 in Catalonia, no obvious outliers appear in
the inventories. The Italian envelope curve lies slightly higher than
the French and Spanish curves, which can be explained by the
much higher coverage of the Italian data base. As illustrated in
Fig. 3 the data sets and calibrated envelope curves appear also to
be consistent in magnitude with previously conducted inventories
on extreme floods in the Mediterranean area (Alcoverro et al.,
1999; Mimikou, 1984; Pardé, 1961) and over the world (Herschy,
2005; Costa, 1987b; Rodier and Roche, 1984; Unesco, 1976). Note
that a limited number reported flood peak discharges over the
world exceed the calibrated Mediterranean envelope curve, espe-
cially for the smaller catchment areas.
A practical conclusion can be drawn from this overall good
agreement between inventories and envelope curves: provided
75
Fig. 3. Comparison between the unit discharges collected for this study, Hydrate
(the dashes) and the unit discharges reported in literature for the world (triangles).
The literature referenced is Costa (1987a,b), Pardé (1961), Rodier and Roche (1984),
Alcoverro et al. (1999) and Mimikou (1984).
that some quality checks are conducted, indirect peak discharge
ð1Þ
estimation methods may provide correct approximate if not accu-
rate discharge estimates.
Flash flood magnitude and occurrence
The data sets mix peak discharge estimates from watershed
areas ranging from some tens of square kilometres to about 2000
km2. According to the adjusted envelope curves and especially
the value of the power b (0.4), a reduced peak discharge Qr has
been selected as an indicator of the magnitude of the floods to
aggregate the data and limit the influence of the watershed areas
on the analyses: Qr = Q/A0.6. Q is in m3/s, A in km2 and Qr in m3/s/
(km2)0.6. The atlas of extreme floods over Europe thus plots the re-
duced discharge of events in each region (Fig. 4). The atlas maps
the spatial distribution of flash floods in the seven areas included
in this study. Differences appear on this map that are confirmed
by examining the proportion of the flash flood events in each re-
duced discharge category for each considered region based on
analysis of the refined data sets (Table 4). There appears to be a sig-
nificant proportion of events of over 75 m3/s/(km2)0.6 in France
(13%), northern Italy (17%) and in Catalonia (11%), and the lower
category events (Qps < 25 m3/s/(km2)0.6) in these regions are a
minority. This lower category represents three quarters of all other
inventories. This reinforces the conclusions drawn on the envelope
curves: the values defining the curves are not isolated in each re-
gion and it is possible to conclude that the differences observed be-
tween regions are not an effect of randomness but correspond to a
real trend.
No significant difference between French and Italian samples
appears upon first inspection of the data. There are, however, some
differences between the distribution of floods in each discharge
category in the samples as there is an over-representation of ex-
treme events in the Italian sample due to the larger coverage, but
a lower number of second class floods that can be explained by a
lower sampling density. There are heterogeneities within the con-
sidered regions as well, especially in northern Italy where the ex-
treme events are concentrated in the Piedmont and Liguria
regions (western area). Likewise, a thorough statistical analysis of
the French sample revealed clear differences in flood magnitude
and occurrence within the considered French Mediterranean area
(Newinger, 2007).
Finally, the flash flood geographical pattern revealed by the
Hydrate data base appears to be correlated to the spatial pattern
of the intense rainfall hazard as shown by the distribution
of estimated 100-year return period daily rainfall accumulations
76 E. Gaume et al. / Journal of Hydrology 367 (2009) 70–78

Fig. 4. Atlas of reduced peak discharges of extreme events in Europe.

Table 4 tumn but also some in spring. This feature is not modified if the
Proportion of events in each reduced peak discharge category in the refined data set complete dataset is taken into account. This difference in the sea-
(%). sonality shows that the most extreme flash floods in the Mediter-
0–25 25–50 50–75 75–100 >100 ranean and in the inland continental regions are not induced by the
Catalonia 45 33 11 11 0
same types of meteorological events. Flash floods in the study area
France 17 43 27 13 0 encompass a diversity of meteorological and hydrological
Italy 33 43 7 10 7 processes.
Slovakia 87 13 0 0 0 Intense autumn storm events delivering very high amounts of
Greece 75 25 0 0 0
rain water within a short period of time and sometimes over large
Romania 77 23 0 0 0
Austria 100 0 0 0 0 areas appear to be a specificity of the Mediterranean area. Intense
summer thunderstorms also take place frequently in this area. But
they almost never induce significant floods except in the arid part
of Catalonia, due to the high infiltration capacities of the dry soils
in the hydrometeorological regions (Fig. 5). This is a sign of in this season. Moreover, even in autumn, daily rainfall accumula-
both, the predominant role played by the rainfall forcing in tions exceeding 200 mm, rarely observed under continental cli-
explaining flash flood hazard and the relevance of the data base. mate, are needed to generate significant floods on vegetated
The detailed analysis of the data nevertheless reveals a more catchments in the Mediterranean area as revealed by recent post
nuanced pattern especially in inner continental countries and flash flood field investigations (Delrieu et al., 2004; Gaume et al.,
particularly the influence of the local meteorological settings 2004; Cosandey, 1993). The initial wetness conditions of the
and of the characteristics of the watersheds (Mertz and Blöschl, watersheds can play a major role in their response to a rainfall
2003). event (Borga and Gaume, 2007). These conditions are linked to
the climatic settings and can also explain the differences observed
Flash flood seasonal distribution between the hydrometeorological regions. In the same line of
thinking, Mertz and Blöschl (2003) found that flash floods in Aus-
The inventory also provides information on the patterns of sea- tria are most extreme in the eastern part of the country, which cor-
sonality of flash floods over the European regions. The number of responds with the flash flood atlas presented here. These authors
records reported for each month is shown for each region on suggest that the hilly terrain in this region increases the instability
Fig. 6. The extreme floods in Catalonia and France, without excep- of the boundary layer and therefore increases the potential for con-
tion, occurred in autumn, with a possible slight shift of the autumn vective storms.
season from year to year, while in the Central European countries Flash flood occurrence and magnitude are controlled by the
(Austria, Slovakia and Romania) these extremes occur only in the combination of meteorological and hydrological factors that flash
late spring and summer seasons. Northern Italy lies in an interme- flood mapping efforts, as the one presented here for Europe, will
diate position with a dominant proportion of extreme floods in au- progressively help to reveal.
 E. Gaume et al. / Journal of Hydrology 367 (2009) 70–78 77

Fig. 5. Distribution of the estimated 100-year return period daily rainfall amounts.

Fig. 6. Number of flash flood occurrences in each month.

78 E. Gaume et al. / Journal of Hydrology 367 (2009) 70–78
Conclusions
Flash flood data from seven hydrometeorological regions in Eur-
ope have been collated. The resulting data base is provided at
http://www.Hydrate.tesaf.unipd.it/ (European flash flood data cen-
ter page). The data set is a step towards a European Flash Flood
database and atlas of extreme events. Events that occurred be-
tween 1946 and 2007 on watersheds of an area less than
500 km2 were collated. Analysis of the data sets shows that flash
floods in each region occur in watersheds of various sizes less than
500 km2. Nevertheless, in Catalonia and France some records of
flash flood events affecting areas greater than 500 km2 had to be
considered as convective storms in the Mediterranean region
may potentially cover wide areas (more than 1000 km2) and cause
intense rainfall over a long duration – up to 12 h. The magnitude of
the flash flood event is measured in this study by reduced peak dis-
charge. The reduced discharges for past floods in Italy, France and
Catalonia have been greater in magnitude than reduced discharges
for Romania, Austria and Slovakia, which suggests that flooding in
the Mediterranean region is generally more extreme than in inland
continental regions. With regard to the time of year during which
the flash floods occur in each region, it is observed from the data
sets that the most extreme flash floods in the Mediterranean region
occur in the autumn months and flash floods in the inland conti-
nental region occur in the summer months, revealing different cli-
matic forcing.
The data catalogue may be developed further by extending the
inventories within the existing regions and also by including other
countries within the data base. With the introduction of the Euro-
pean Floods Directive in 2007 requiring EU member states to have
prepared flood risk maps by 2013, there is likely to be an increase
in efforts to research and document historic floods as part of the
work done to meet the Directive. This may enable access to data
that was not readily available to this study. In addition to further
developing the data catalogue, it may be interesting to further de-
velop the analyses of the data. This may include statistical ap-
proaches to drawing inter-comparisons of the data especially to
reveal the heterogeneity of the data between and within the con-
sidered regions and to evaluate the return period of the reported
extreme events.
Acknowledgements
The research presented in this paper has been carried out as
part of the European Commission funded project, Hydrate, Sixth
Framework Programme, Contract No. 037024.
References
Alcoverro, J., Corominas, J., Gómez, M., 1999. The Barranco de Arás flood of 7 August
1996 (Biescas, Central Pyrenees, Spain). Engineering Geology 51, 237–255.
Anselmo, V., 1985. Massime portate osservate o indirettamente valutate nei corsi
d’acqua subalpini. Atti e Rassegna Technica Societa Ingegneri e Architetti in
Torino – Nupva Serie 39 (10–12), 245–275.
Barredo, J.I., 2007. Major flood disasters in Europe: 1950–2005. Natural Hazards 42,
125–148.
Bayazit, M., Onoz, B., 2004. Envelope curves for maximum floods in Turkey. Digest
2004, 927–931.
Borga, M., Gaume, E. (2007). Hydrological processes and flash flood generation: new
observations and concepts. In: Proceedings of the 32nd IARH congress, Venice.
Castellarin, A., 2007. Probabilistic envelope curves for design flood estimation at
ungauged sites. Water Resources Research 43, W04406. doi:10.1029/
2005WR004384.
Cosandey, C., 1993. La crue du 22 septembre 1992 sur le Mont Lozère. Revue de
Géomorphologie Dynamique (2), 49–56.
Costa, J.E., 1987a. Hydraulics and basin morphometry of the largest flash floods in
the conterminous United States. Journal of Hydrology 93, 313–338.
Costa, J.E., 1987b. A comparison of the largest rainfall–runoff floods in the United
States with those of the People’s Republic of China and the World. Journal of
Hydrology 69, 101–115.
Costa, J.E., 1988. Rheologic, geomorphic, and sedimentologic differentiation of
water floods, hyperconcentrated flows, and debris flows. In: Baker, V.R., Kochel,
R.C., Patton, P.C. (Eds.), Flood Geomorphology, 7. John Wiley & Sons, Chichester,
England, pp. 113–122.
Creutin, J.D., Borga, M., 2003. Radar hydrology modifies the monitoring of flash
flood hazard. Hydrological Processes 17 (7), 1453–1456. doi:10.1002/
hyp.5122.
Crippen, J.R., Bue, C.D., 1977. Maximum flood flows in the conterminous United
States. US Geological Survey Water Supply Paper, 1887.
Delrieu, G., Ducrocq, V., Gaume, E., Nicol, J., Payrastre, O., Yates, E., Andrieu, H.,
Ayral, P.-A., Bouvier, C., Creutin, J.-D., Livet, M., Anquetin, S., Lang, M., Neppel,
L., Obled, C., du Châtelet, J. Parent., Saulnier, G.-M., Walpersdorf, A., Wobrock,
W., 2004. The catastrophic flash-flood event of 8–9 september 2002 in the
Grad region, France: a first case study for the Cévennes-Vivarais
Mediterranean hydro-meteorological observatory. Journal of
Hydrometeorology 6, 34–52.
Dobbie, C.H., Wolf, P.O., 1953. The lynmouth flood of August 1952. Proceedings of
the Institute of Civil Engineers 2, 522–588.
Ferro, G., 2005. Assessment of major and minor events that occurred in Italy during
the last century using a disaster severity scale score. Prehospital and Disaster
Medicine 20 (5), 316–323.
Gaume, E., Livet, M., Desbordes, M., Villeneuve, J.-P., 2004. Hydrological analysis of
the river Aude, France, flash flood on 12 and 13 November 1999. Journal of
Hydrology 286, 135–154.
Guzzetti, F., Stark, C.P., Salvati, P., 2005. Evaluation of flood and landslide risk to the
population in Italy. Environmental Management 36 (1), 15–36. doi:10.1007/
s00267-003-0257-1.
Herschy, R.W., 2005. The world’s maximum observed floods. Flow Measurement
and Instrumentation 13, 231–235.
Huet, Ph., Martin, X., Prime, J.-L., Foin, P., Laurain, Cl., Cannard, Ph., 2003. Retour
d’expériences des crues de septembre 2002 dans les départements du Gard, de
l’Hérault, du Vaucluse, des bouches du Rhône, de l’Ardèche et de la
Drôme.Inspection générale de l’Environnement. Paris, France. 124 p.
Jarrett, R.D., 1987. Errors in slope-area computations of peak discharges in
mountain streams. Journal of Hydrology 96, 53–67.
Jarvis, C.S., 1924. Flood flow characteristics. Proceedings of the American Society of
Civil Engineers, 1545–1581.
Kadoya, M., 1992. Study on record flood peaks in Japan. Proceedings of the Japan
Academy, Series B 68 (8), 133–138.
Lefrou, Cl., Martin, X., Labarthe, J.-P., Varret, J., Mazière, B., Tordjman, R., Feunteun,
R., 2000. Les crues des 11, 12 et 13 novembre 1999, dans les départements de
l’Aude, l’Hérault, les Pyrennées Orientales et du Tarn. Inspection générale de
l’Environnement. Paris, France. 140 p.
López Bustos, A., 1964. Resumen y conclusiones de los estudios sobre avenidas del
Valles en 1962. Instituto de Hidrología, Technical Report, Madrid.
Mertz, R., Blöschl, G., 2003. A process typology of regional floods. Water Resources
Research 39 (12), 1340. doi:10.1029/2002WR001952.
Marchetti, G., 1953. Sulle massime portate di piena osservate nei corsi d’acqua
Italiani a tutto il 1953. Giornale del Genio Civile 3 (4), 193–206.
Matthai, H.F., 1969. Floods of June 1965 in South Platte River Basin Colorado. Water
Supply Paper 1850 – B. US Geological Survey, Washington DC.
Mimikou, 1984. Envelope curves for extreme flood events in north-western and
western Greece. Journal of Hydrology 67, 55–66.
Münich Re Group, 2003. Annual review: natural catastrophes 2002. Münich Re
Group, Münich, Germany. 62 p.
Newinger, O., 2007. Inventaire et analyse des crues éclair du Sud de la France
de 1950 à 2007. Master thesis. Université Pierre et Marie Curie, Paris,
France.
Ning, C., Zhaohui, W., 1986. A critical review of research on the hyperconcentrated
flow in China. Series of Publication IRTCES, Beijing.
Pardé, M., 1961. Sur la puissance des crues en diverses parties du monde.
Geographica 8, 1–293.
Pierson, T.C., Costa, J.E., 1987. A rheologic classification of subaerial sediment–water
flows. In: Costa, J.E., Wieczorek, G.F. (Eds.), Debris Flows/Avalanches: Process,
Recognition, and Mitigation Reviews in Engineering Geology, vol. 7. Geological
Society of America, pp. 1–12.
Rodier, J.A., Roche, M., 1984. World Catalogue of Maximum Observed Floods. IASH
Publication Number 143. IASH Press.
Stanescu, V.A., 2000. The Extreme Floods in Europe: An Analysis of Their Occurrence
and Regionalization. Unpublished note.
Stanescu, V.A., 2004. Regional analysis of annual peak discharges in the Danube
Catchment. The Danube and its catchment—A hydrological Monograph, Follow-
up volume VII. Regional Co-operation of the Danube Countries in the Frame of
the International Hydrological Programme of UNESCO, Bucharest.
Unesco, 1976. World Catalogue of very Large Floods. The Unesco Press.